CreateTableQuery: use correct VARCHAR length

[bevzzz]

This PR fixes the handling of VARCHAR length in CREATE TABLE, see #736.

Changes:

• Dialect implementations now need to have DefaultVarcharLen() uint, so that we wouldn't need to hardcode the default lengths in the dialect-specific types
• Minor refactor of CreateAlterTable.appendSQLType
• Since VARCHAR can only have positive length, the signature of CreateAlterTable.Varchar is changed to Varchar(n uint).
  This commit , therefore, has a BREAKING CHANGE ❗, as this would no longer compile:

var n int = 4
db.NewCreateTable().Model(...).Varchar(n)

Edit: breaking changes were reverted, see this commit instead.

This change is not necessary for the fix, I only propose this for the sake of additional "type safety" in the queries. It feels more transparent than something like n = max(0, n). However, if such API change is not acceptable, we could also return an error for q.varchar <= 0 or just revert this commit entirely.

[vmihailenco]

@bevzzz looks good, thanks.

Could you clarify what happens when a dialect returns varchar=0 and CreateTableQuery does not specify Varchar length either? Do we have a test in query_test for such a case?

However, if such API change is not acceptable, we could also return an error for q.varchar <= 0 or just revert this commit entirely.

Let's not break the existing API. Instead, let's return an error like you suggested or just panic if returning an error is complicated.

Also, len(slice) can return only positive values too, but the return type is int, not uint. Just saying :)

[bevzzz]

what happens when a dialect returns varchar=0 and CreateTableQuery does not specify Varchar length either?

We initialize q.varchar with the dialect's default length (see here), so if no other Varchar is specified (q.varchar == 0), we do not append anything in the parenthesis and the type stays VARCHAR .

Yes, there's a test case for that:

func(db *bun.DB) schema.QueryAppender {
    return db.NewCreateTable().Model(new(Model))
}

It's one of the older ones (no regressions ✌️ ), and for PG, which has default 0, it generates this query:

CREATE TABLE "models" ("id" BIGSERIAL NOT NULL, "str" VARCHAR, PRIMARY KEY ("id"))

Let's not break the existing API

Agreed. Good point about len(slice) too👍

[vmihailenco]

Thanks 👍